Google's new Gemini 4 Hydrogen understands the most muddled and unclear requests

14/11/2026, 17:282,201
1 / 2

Google says its new AI model Gemini 4 Hydrogen is better able to understand natural speech, complex wording and even requests containing typos.

A user can ask several questions at once, jump from one thought to another or phrase a request less than clearly, and the neural network will try to work out their intention without asking for further clarification.

In the Gemini app, the model takes better account of the context of lengthy conversations and more accurately follows a specified style and format when writing emails, editing texts and preparing documents. It can also use data from Gmail, Google Drive, Airtable and Dropbox for multi-step tasks.

Gemini 4 Hydrogen has four reasoning modes, ranging from low to maximum. At the standard processing depth, according to Google, it runs 2.5 times faster than its predecessor, Gemini 3.8 Flash, and the number of fabricated statements has fallen more than fivefold. In internal testing of AI agent honesty, it scored 96.7%.

The model can analyse PDF files, spreadsheets, charts, audio and video, find errors in code and produce reports and presentations based on this data. Its context window reaches one million tokens, as does the maximum length of a generated response.

Gemini 4 Hydrogen is already being rolled out worldwide. In addition to Gemini and Google Search, it is available to developers through Google AI Studio and the Gemini API. Google AI Pro and Ultra subscribers can also try it out in Google Antigravity and Google Sheets.

Disclaimer: this article was translated by artificial intelligence and may contain inaccuracies.